How It Works

The critical difference is the largest mean difference between the new and the current lot that would not change how a result is used. It is the share of the analyte's allowable total error (TEa) allowed for a change of reagent lot, set for each level in one of three ways:

• TEa − 2 × within-lot SD (the default) — leaves room for the everyday random error of a single lot and gives the rest to the lot change.
• Percent of TEa — a share you choose, e.g. 50% of a 6 mg/dL TEa gives 3 mg/dL.
• Enter a value — a limit agreed with clinicians, in the analyte's units or as a percentage.

The TEa comes from the analyte in the test directory. When a critical difference is based on the TEa, Data Entry shows it under Total Allowable Error, where you can override it for this comparison only (the test directory is not changed); a plan can set the same override. It is not shown when every level's critical difference is entered. Data Entry shows the resulting critical difference with its working.

Power is the chance that the comparison catches a new lot whose true mean difference equals the critical difference. A power of 90% means a lot that is off by exactly the critical difference is rejected nine times out of ten; lots that are off by more are rejected even more often. Higher power needs more samples. A good lot is wrongly rejected no more than 5% of the time, whatever power you choose.

Choose 90% or 95% when a small change in a single result can change a treatment decision, and 80% when results are rarely used alone.

How many samples are needed depends on how much the results scatter on a single lot. Two figures are used for each level: repeatability (within-run scatter) and within-lot imprecision (scatter across runs and days on one lot). Legion Diagnostics fills both in from the package-insert precision claims for the analyte. They sit in the collapsed Imprecision section of Data Entry, where you can check them or change them for one comparison; to change the claim itself, edit it in the Test Directory.

From the critical difference, the power and the imprecision, Legion Diagnostics suggests how many patient samples to test at each level (never more than 40) and works out the rejection limit for the number you actually test. The rejection limit is a fraction of the critical difference: the observed mean difference must stay below it for the new lot to be accepted. Using a limit tighter than the critical difference is what lets a handful of samples reach the chosen power.